FAQ

What is RealSim-CP, in one sentence?

A high-fidelity, multimodal (camera + LiDAR) cooperative-perception dataset for Japanese urban traffic, generated with the physics-based DIVP simulator and annotated in ASAM OpenLABEL 1.0.0.

Is the data real or simulated?

Simulated — but with a physics-based simulator (DIVP) that ray-traces sensor operation and is calibrated against real measurements, so it stays close to real sensor behavior. See Simulator (DIVP).

How do I download it?

From the project's Google Drive — see Download. You can grab a single scenario/variant (a few GB) rather than everything.

How is it licensed? Can I use it commercially?

The dataset is CC BY 4.0 — yes, including commercial use, with attribution (cite the paper). The code/tools are MIT. See DATA_LICENSE.md and LICENSE.

What annotation format is used?

ASAM OpenLABEL 1.0.0 — one JSON file per variant with streams, the coordinate-system tree, per-frame poses, and 3D object cuboids. See Label format.

Why does the agent pose not match the vehicle body center?

The vehicle_*_local frame is the parent for mounted sensors and sits near the rear axle, not the body center. Use the matching labeled cuboid for the body. See Label format.

Some daytime scenes are missing point clouds — is that a bug?

The Daiba sunny_1 / sunny_2 variants have incomplete or absent LiDAR. Use a night variant for full LiDAR coverage, and make your loader tolerate missing files. See Known issues.

What sensors are simulated?

Cameras modeled on the Sony IMX490 (120° HFOV) and LiDAR modeled on the Velodyne VLS-128, at 10 Hz. See Sensors.

How do I view a scene quickly?

Install the Visualizer (Python 3.12 + Open3D) and run python -m realsim_viz <variant_path>. See the Quickstart.

How do I cite RealSim-CP?

See the Citation section of the README or CITATION.cff.

How do I report a problem or ask a question?

Open an issue on GitHub.